Date: Wed, 23 Oct 2002 11:26:49 0500
ReplyTo: "Simon, Steve, PhD" <ssimon@cmh.edu>
Sender: "SPSSX(r) Discussion" <SPSSXL@LISTSERV.UGA.EDU>
From: "Simon, Steve, PhD" <ssimon@cmh.edu>
Subject: Re: design question (Non SPSS)
ContentType: text/plain
Tony Baglioni writes:
> I've forgotten so many basics. Can anyone tell me how to
> calculate the needed sample size for a simple query? I
> know the probability of an event is 3/10,000. What sample
> size is needed to verify there are no cases present?
You need to be a bit more precise in your research hypothesis. If you
already know the probability of an event, why are you running a test at all?
I'll make a guess that you want to find a sample large enough so that you
can exclude the possibility that the true proportion is 3/10,000. In other
words, you want to prove that the true proportion is less than 3/10,000.
The rule of three is helpful here. If you observe zero events out of n, then
3/n is an approximate 95% confidence interval for the true probability.
Solve the equation
3/n < 0.0003
to get n > 10,000.
A more formal power calculation could be done, and it would probably demand
an even larger sample size. You need to be more explicit with your
hypothesis before you can do a power calculation. I understand that StatXact
will do a power calculation for you, although I do not have the latest
version which does this. The traditional formulas for power fall apart
because the event is very rare.
Steve Simon, ssimon@cmh.edu, Standard Disclaimer.
The STATS web page has moved to
http://www.childrensmercy.org/stats.
